The impact of introducing an interest barrier - Evidence from the German corporation tax reform 2008

Transcription

1 The impact of introducing an interest barrier - Evidence from the German corporation tax reform 2008 Hermann Buslei DIW Berlin Martin Simmler 1 DIW Berlin February 15, 2012 Abstract: In this study we investigate the impact of the thin capitalization rule newly introduced in Germany in 2008 on firms capital structure, investment and profitability. The identification of the causal effects is based on the escape clauses within the regulation using a difference-in-difference approach. Our results suggest that firms strongly react to the new regulation in order to avoid the limited deductibility of interest expenses either by decreasing their debt ratio or by splitting the firms assets to use the exemption limit. We further show that the effect on firms investment depends on the firms financial situation. Thus, negative investment effects are not found in general. With respect to the aim of restricting the profit shifting of multinational firms our results indicate that the sensitivity of investment to corporate income taxes increases for firms, which are affected by the thin capitalization rule. Moreover, our analysis shows that the newly introduced thin capitalization rule is quite successful in broadening the tax base. Keywords: thin capitalization, earnings stripping rule, debt ratio, profitability, investment. JEL Classification: H25, H26, G32. 1 Corresponding author, German Institute for Economic Research (DIW Berlin), Berlin, Germany, phone: , fax: , msimmler@diw.de

2 1. Introduction Profit shifting is a severe problem for governments in Europe and the US as prior studies have shown (Huizinga and Laeven, 2008, Weichenrieder, 2009). While the growing international tax competition has led to a general decline of the statutory corporate tax rates, several countries especially the larger ones have additionally introduced or tightened thin capitalization rules (see Haufler and Runkel, 2008). The main objective of these rules is to prevent firms from shifting profits abroad. How effective thin capitalization rules are and which behavioral adjustments they provoke, however, has not yet been analyzed in great detail. Buettner et al. (2008) analyzed the impact of thin capitalization rules using a firm level panel data set of the affiliates of German multinationals in 36 countries combined with information on corporate taxation in each of these countries. They find that thin capitalization rules reduce the amount of intercompany loans (internal debt of firms). Additionally, they show that imposing a thin capitalization rule decreases the tax sensitivity of internal debt. Firms for which a thin capitalization rule is binding show a weaker reaction to tax rate changes. Further, the authors report evidence that there is an overall reduction in investment as a response to thin capitalization rules. Overesch and Wamser (2010) show that the German thin capitalization rule in the period 1996 to 2004 reduced internal borrowing significantly. However, Wamser (2008) reported in his study that the reduction of internal borrowing came along with an increase in external debt. All these studies did not consider the profitability of the firms affected by the thin capitalization rule and thus the impact of thin capitalization rules on the tax base. To fill this gap and to give a comprehensive picture of the effects of thin capitalization rules is the aim of this study. To do so, we analyze first, how the capital structure (taking into account internal and external debt) of firms was affected by the new thin capitalization rule introduced in Germany in the year In a second step, we study the impact of the thin 1

3 capitalization rule on investment. Finally, we investigate how effective the newly introduced thin capitalization rule was with respect to the aim of broadening the tax base. 2 The thin capitalization rule introduced in Germany in 2008 incorporates the experience from the old regulation, where only payments to direct shareholders had been considered. The disadvantages of the regulation were on the one hand that firms substituted external and internal debt as shown by Wamser (2008) and on the other hand that it was hard to distinguish between shareholders and third parties as creditors in complex group structures. Thus, under the new regime, interest expenses (independent of the creditor) may only be fully deducted if the net interest expenses do not exceed 30 % of EBITDA, adjusted for tax purposes. Therefore, in principle all firms in Germany are affected by the limited interest deductibility. However, as the German government was not interested in harming its own economy by implementing this broad concept of a thin capitalization rule it included several escape clauses as well. The most important one for the majority of firms is the tax exemption limit for the net interest result of 1 million euro 3. This is also the regulation on which we base the identification strategy in this paper to analyze our research questions. Our analysis uses a nearly complete set of financial statements data for all incorporated German firms. Methodically, we use a difference-in-difference approach to identify the causal effect of the thin capitalization rule. Our results suggest that firms strongly react to the new regulation in order to avoid the limited deductibility of interest expenses either by decreasing their debt ratio or by splitting the firms assets to use the exemption threshold. We further show that the effect on firms investment depends on the firms financial situation. Thus, negative investment effects are not found in general. With respect to the aim of restricting the profit shifting of multinational firms our results indicate that the sensitivity of investment to corporate income taxes increases for firms, which are affected by the thin capitalization rule. Moreover, our analysis shows that the newly introduced thin capitalization rule is quite successfully in broadening the tax base. 2 Note that although financing decisions and the tax base are in principle positively related, the changes in both items as a reaction to a thin capitalization rule need not be identical. For example, the tax base might increase less than equity if firms try to avoid the expected increase in the tax burden by relying more on other profit shifting opportunities, for example the choice of transfer prices. 3 The threshold of 1 million euro has been increased by the peoples' relief act (Bürgerentlastungsgesetz) (temporarily) and the Growth-Enhancement Act ("Wachstumsbeschleunigungsgesetz") (permanently) in the year 2009 to 3 million euro. However this was after December 31,

4 The outline of the paper is as follows. In the next section we provide a summary of the new thin capitalization regulation in Germany. Section three introduces the methodology used in this study, afterwards we present our dataset. Results are reported and discussed in section 5, whereas section 6 concludes. 2. Institutional Background In the coalition contract signed by the so called large coalition of Christian democrats and Social democrats on November 11, 2005, the coalition partners announced a major corporation tax reform. A first draft bill for this reform ("Unternehmensteuerreform 2008") was presented by the German government on February 5, 2007 (see Rödder and Stangl, p. 479). The law passed the last stage of the legislation process (Bundesrat) on July 6, 2007 and was published on August 14, Those parts relevant for this study were enacted at the beginning of the year One of the important elements of the corporation tax reform 2008 was the introduction of a new interest barrier which limits the deductibility of interest payments under certain conditions and is in several characteristics more restrictive than the regulations that preceded it. The new interest barrier takes into account interest payments from all types of creditors and applies to all types of companies. Before the reform in the year 2008, interest payments were generally deductible from total revenues as regular expenses. The only major exception from this rule was made for interest payments on credits provided by shareholders of limited liability companies who's share in the company exceeded a certain threshold (article 8a German Income Tax Code in the year 2007 and earlier). These interest payments were regarded as masked (covered) profit transfers and were qualified as hidden dividend payments and taxed respectively. Obviously, the government did not regard the old regulation as being effective in preventing companies from shifting profits to lower taxing countries (see Rödder and Stangl, p. 479, Deutscher Bundestag, Drucksache 16/4841, p. 29). The main cause for the ineffectiveness of the old regulation might have been the problem to 4 The probably most important change of the reform was the reduction in the tax rate of the corporation tax from 25 to 15\%. Gradual changes of the law which are of interest here were included in the peoples' relief act (Bürgerentlastungsgesetz) and the Growth-Enhancement Act ("Wachstumsbeschleunigungsgesetz"). The former passed the Bundesrat on July 10, 2009 and the latter on December 18,

5 differentiate between financing of owners and third parties in those cases where a relation between them seemed to be present but could not be proved (Thiel, 2007). The basic rule of the new regulation states that interest payments are deductible as long as they are balanced by interest income or in case interest payments exceed firms interest income as long as the exceeding payments are less or equal to 30 % of earnings before interest, taxes, depreciation and amortisation (EBITDA) (see Art. 4h German Income Tax Code). The new regulation applies to the corporate income, the local business and also to the income tax. Interest payments which are not deductible in one year may be carried forward indefinitely and - given sufficiently high levels of EBITDA in later years - may be deducted then. 5 In order to prevent small firms and firms with a somehow "sufficient" equity financing from an additional tax burden, the interest barrier comes with several "escape clauses". Small companies should not face a burden as already the initial code in the year 2008 provided a tax exemption limit (Freigrenze) of 1 million euros. 6 The second escape clause applies to companies which do not belong to a group do and do not rely on significant shareholder debt financing (stand-alone-escape, Art. 8a, 2 Corporation Tax Code). The standalone escape may also apply to consolidated tax groups. If all members of a group form a single tax group, the whole group is regarded (and treated in the same way) as a single company. The whole group is thus exempted from the interest barrier in case that no harmful financing by owners is present. The third escape type is provided for members of a group which does not qualify as a tax group. A group member may deduct all interest payments if the member's equity rate (equity over total assets) is not lower than the equity rate in the whole group (according to the consolidated statement of the group including the company under consideration). The 5 Excess interest payments are not re-qualified as hidden dividend payments as was the case with the earlier interest barrier which applied to affiliates only. Thus, for foreign creditors whose interest income is taxable at home, the new legislation leads to a double taxation of the interest payments. As stated in the explanatory statement to accompany the new legislation, foreign creditors should be given an incentive to finance investments in Germany with equity. While paying the tax on the excess interest would be equal to paying the tax on dividends of the same amount in Germany, the corporations would avoid the double taxation in the second case as dividend payments from abroad are usually tax free in the home country of the investor. 6 This limit was first increased to 3 million euro in the peoples' relief act (Bürgerentlastungsgesetz) in the year 2009 for business years which started on May 25, 2007 the earliest and ended on January 1, The Growth-Enhancement Act ("Wachstumsbeschleunigungsgesetz") suspended the time limitation of the increased tax exemption limit. 4

6 initial regulation of the corporation tax reform 2008 included a tolerance level of 1 percentage point. 7 Like the second escape type, also this third type is granted only for member companies of groups if limits for shareholder debt financing are not violated by any member in the whole group (Art. 8a, 2 Corporation Tax Code). 8 Due to the specific design of the interest barrier, especially the design of the escape clauses, adjustments to the interest barrier might be extremely complex. This is most obvious for large corporations with a huge number of shareholders and/or subsidiaries in Germany and/or in foreign countries. The adjustment might result in shifting of equity between all members of a group. This could be the case if the ultimate owner tries to keep the interest payments of the group members below the exemption limit or tries to fulfill the conditions of the equity escape clause. Without precise knowledge of the relevant information for all group members, not even the sign of the change in equity for a group member can be predicted. The adjustments might also lead to major revisions of the group structure, for example by including or excluding companies from a tax group. It hardly seems to be possible to capture these options for all firms in publicly available data sets. Therefore, we restrict the investigation of the reactions to the interest barrier on companies for which it seems to be a realistic option to reduce their interest payments below the tax exemption limit. For these companies (or their owners) it is likely that using the exemption limit is their option of first choice. In this case, adjustments of equity dominated by the demands of the whole group are probably less severe than for companies for which interest payments are much higher than the exemption limit. 9 7 This level was raised to 2 percentage points in the Growth-Enhancement Act in Probably as a reaction to the criticism during the economic downturn in the year 2009, the Growth- Enhancement Act ("Wachstumsbeschleunigungsgesetz") in the year 2010 introduced an EBITDA carry forward. As mentioned above, the basic rule of the interest barrier states that net interest payments up to 30 % of current EBITDA are deductible. Under the new legislation, companies with interest payments below 30 % of EBITDA in a calendar are granted an EBITDA carry forward. The amount of the carry forward is equal to the difference between 30 % of EBITDA and the net interest payments. The EBITDA carry forward may be used in the five following years and is simply added to the value of current EBITDA (see Rödding, 2009). 9 Nonetheless, they might be present and might in part of the cases prevent companies from increasing their equity in order to keep their interest payments below the tax exemption limit. 5

7 Since the regulation described above was not executed as it has been changed in 2009 retrospectively we only include observation up to the year 2008 in our analysis as in these years firms behaved as the regulation would apply Methodology To analyze how the introduction of the thin capitalization rule affects the financing and investment decision of firms and their profitability we use a difference-in-difference approach, which is derived from the evaluation literature. 11 To get unbiased coefficient using a difference-in-difference approach three important requirements have to be fulfilled. Firstly, treatment and control group should have had the same trend before the treatment; we show in the data section that this requirement is met. The second requirement is that there are no other cofounding treatments, although this cannot be tested formally, we carefully form treatment and control group such that other confounding treatments can be ruled out. The last important requirement is that the treatment is exogenous. This we ensure by constructing treatment and control group based on the firm characteristics of 2006 as in this year the regulation of the new thin capitalization rule was unknown. 12 Our treatment in this study is whether the net interest expenses of a firm exceed 1 million euro since in this case the thin capitalization rule would in principle apply. To ensure exogenous treatment, we used the firm characteristics of the year 2006 to form treatment and control group. For two reasons we had to restrict our sample. On the hand, we do so to ensure that the common trend assumption for treatment and control group is fulfilled. On the other hand we do so to ensure that for the firms included in the analysis it can reasonably be assumed that a reduction of their interest payments below the exemption limit is their option of first choice in order to avoid the interest barrier. Thus, we do not use all observations above and below the cutoff point but only those within the range of net interest expenses in 2006 of 500 thousand euro and 1,500 thousand euro. In addition, to avoid misclassification at the threshold we exclude firms with net interest expenses between 10 The most important changes of the regulation in 2008, which apply also for the year 2008, was that the exemption threshold was increased to 3 million euro and that the unused amount of deductible interest expenses based on the ratio of 30% of net interest expenses to EBITDA could be carry forward. 11 See Meyer (1995). 12 See section 2. 6

8 800 thousand and 1,200 thousand euro. Our control group therefore consists of firms with net interest expenses in 2006 between 500 thousand and 800 thousand euros, our treatment group shows expenses between 1,200 and 1,500 thousand euro. Since firms can be assumed to be forward looking, we also exclude all observations in The first question that we analyze is the impact of the new interest barrier on firms debt ratio, defined as liabilities to total assets (DR). Our econometric specification of this relationship is shown in equation (1). Note that estimation is done in differences between 2008 and 2006 to account for firm-specific effects, although they should not matter given the exogenous treatment. The dependent variable in this equation is the change in the firms debt ratio between the years 2008 and 2006 ( ). Treatment is a dummy variable which is one for the treated firms and After is a dummy variable which is one for years after is an i.i.d error term. All other factors which determine firms debt ratio are captured in the vector X. These variables are in our study besides the tax rate on business income firm size, firm age, firms share of tangible assets and the ratio of EBITDA to total assets. The construction of the variables is described in section 4. Please note that we do not have to control for differences between the treatment and control group in equation (1), since we estimate in differences and construct the two groups based on the firm characteristics in The treatment effect is given by. (1) As the thin capitalization rule introduced in 2008 does not only include the exemption limit of 1 million euro for the net interest result but also other escape clauses (see section 2), it is likely that the coefficient γ in equation (1) is biased. Thus, we model the most important other escape clauses in order to identify the effect of the regulation. These are the EBITDA escape, the stand-alone and the tax group stand-alone escape clause. We construct dummy variables for each of the modeled escape clauses and interact them with the Treatment*After and the After variable. The dummies for the entitlement of another escape clause are modeled in such a way that the treatment effect is still given by γ. Since we model the entitlement of the escape clause based on the characteristics in 2006 and estimate in differences, we do not have to control for differences between the firms which 7

9 are entitled and which not. The equation we estimate is exemplary given for one other escape clause in (2): (2) The same econometric approach as described above is used for the analysis of the impact of the interest barrier on investment and on the tax base of a firm (profitability). The construction of the variables follows in the next section, where we also describe the data set on which our study is based. 4. Data The database for our study is the financial statements collection DAFNE provided for German firms by Bureau van Dijk. The main source for this data base is the registrar of companies in Germany. The dataset contains individual balance sheets, profit and loss accounts, and information on ownership structures. For years after 2005 the database covers nearly all incorporated firms in Germany as for these firms strict publication requirements apply. 13 For unincorporated business the database is only representative for unlimited partnerships with a limited liability company as general partner (GmbH & Co. KG). However, as we are interested in firms with net interest expenses above 500 thousand euros (see section 3) and probably only few partnerships with unlimited liability have interest payments above this amount the insufficient representation of these firms in our data base should not have a severe impact on the results of our empirical analysis. From the description of the rules of the interest barrier given above, it is obvious that the information on the net interest result is crucial for the analysis. The net interest result can generally be calculated based on the information in DAFNE, however the relevant information is directly observed only for a subsample of the data. The reason is that the disclosure rules are less strict for the income statement than for the balance sheet. Small 13 In principle all German incorporated companies have to publish their financial statements according to Art. 325 Commercial Code, only subsidiaries companies which fulfill special requirement (see Art. 264 III Commercial Code) are not obliged to do so. To the best of our knowledge only a few thousand companies fulfill these requirements. 8

10 companies are not legally liable to publish the income statement at all. 14 In the DAFNE wave which we use for our empirical study (wave 174 from August 2011), we observe in total around 870 (940) thousand companies with a valid information for the balance sheet (total assets available) in the year 2006 (2008). 15 From these companies, around 100 (90) thousand also provide an income statement and thus we directly observe the information on the net interest result. For only a subsample of them we also have information on the ownership structure which is important to determine which firms are entitled to escape clauses of the thin capitalization rule. Since the selection of companies with an income statement and the selection of companies with the necessary ownership information are probably nonrandom, we construct three different estimation samples to check whether sample selection drives our results and to what extent conclusions can be drawn for the whole population. In the first sample, we only include observations for which an income statement and information on the ownership structure is given. In the second sample, we include all firms for which we observe an income statement and in the third sample, we include all firms. To construct control and treatment group in the latter sample, we impute net interest expenses in 2006 using the observed firm balance sheet characteristics The criteria for size are total assets, sales and number of employees. Small companies fulfill at least two of the following three conditions: 1. total assets are equal or less than million euros, sales are equal or less than million euros and the number of employees is equal to or less than 50. A medium sized company does not fulfill at least two of the conditions which qualify for a small company and does fulfill at least two of the following three conditions: 1. total assets are equal to or less than million euros, sales do not exceed million euros and the number of employees does not exceed 250. For a large company, at least two of the values for assets, sales and employees exceed the respective thresholds for a medium sized company. Moreover all companies listed at an organized bond market are considered as large companies. See article 267 of the German commercial code. 15 All numbers refer to companies which non-consolidated statements. Companies for which only a consolidated statement is available are neglected here and in the empirical analysis. 16 The imputation was done with an OLS regression where the following covariates have been included: Unpaid contributions on subscribed capital, fixed assets, assets in between fixed and current assets, current assets, equity, special item with an equity portion, accruals, liabilities, deferred income (all scaled by the book value of total assets), intangible assets, tangible assets, financial assets (all scaled by the book value of fixed assets), inventories, receivables and other assets, securities, cash-in-hand (all scaled by the book value of current assets), liabilities up to one year, liabilities with a majority of more than one year, loans, liabilities to banks, payments on account of orders, trade payables, liabilities from central settlement, liabilities on bills accepted and drawn, liabilities to shareholders, payable to affiliated enterprises, payable to enterprises in which participation are held, other liabilities (all scaled by the book value of liabilities). We further include the log of total assets as well as legal form and industry dummies. The R 2 of the regression amounts to The results are not shown but available upon request from the authors. 9

11 Noteworthy, from all samples, we excluded financial firms as well as firms within the sectors public administration and defense, education, health and social work and other community activities and firms with negative equity. The number of observations in sample 1 (2 and 3) amounts to 704 (1,090 and 1,728), from which 168 (270 and 447) belong to the treatment group. We turn to the construction of variables next. For our first research question the dependent variable is the change in the debt ratio between 2008 and We follow the literature and define the debt ratio as the ratio of liabilities to the book value of total assets. Besides the interaction term between the dummy indicating treatment and the dummy indicating the year after the reform, we include the following other determinants as covariates for the change in the debt ratio as dependent variable: change in firm size (log. of total assets in thousand euro), log firm age, firms share of tangible assets (ratio of tangible assets to book value of total assets) as well the ratio of EBITDA to the book value of total assets. Further, we include the tax rate on business income to control for changes due to the German corporate tax reform in For incorporated firms the tax rate on business income captures the corporate income tax and the trade tax. 17 For incorporated business the tax rate depends on the shareholder structure as these firms divide their income among the shareholders and pass it through to the shareholders. For non-natural persons as shareholder the tax rate captures the tax rate on corporate income and the trade tax; for natural persons it is the tax rate on business income plus trade tax. 18 Descriptive statistics of the variables used in the estimation for the whole sample (sample 1) and for treatment and control group are shown in table 1. On average a firm in sample 1 has in 2006 a debt ratio of 64 %, an investment rate of 11 % and a ratio of profit to total assets of 4.8 %. Debt ratio and investment rate do not differ between treatment and control group. The profitability of firms in the control group is, however, higher than for treated firms (5.2% to 4.4%). With respect to the control variables, treatment and control group differ significantly in their firm size and in their share of tangible assets. Further, both 17 We obtained the local trade tax rates by merging the local business tax rates provided by the Statistical Offices ( ) to the database by using the firms postal codes provided in Dafne. 18 In case the ownership structure of a partnership firm is not observed, we used imputed average ownership structure. 10

12 groups operate partly in different industries. Descriptive statistics for sample 2 and sample 3 are given in Table A-1 in the Appendix A. Table 1: Firm characteristics 2006 of treatment and control group (Sample 1) Full Sample Control Group Treatment Group Variable N Mean Std dev. Mean Mean p-value liabilities (thd. ) interest rate debt ratio investment quota Profitability t-test firm age ,969 2,947 0,81 tangibility 704 0,514 0,284 0,492 0,585 0,00 firm size (thd. ) , corporate tax rate 704 0,396 0,031 0,397 0,395 0,42 Industries (shares): manufacturing 704 0,338 0,473 0,351 0,298 0,21 trade 704 0,229 0,420 0,246 0,173 0,05 services 704 0,051 0,220 0,043 0,077 0,08 Notes: Statistics are for Source: DAFNE firm database, 2006, own calculations. Since the exemption threshold is not the only escape clause included in the newly introduced interest barrier, we modeled - where possible - also the other escape clauses in order to account for them in the estimation. Given our data, we are able to consider (to som extend) the stand-alone-clause, the tax group stand-alone-clause and the EBITDA escape clause. The entitlement of each escape clause was modeled as follow: Stand-Alone-Clause: In principle a firm is considered as a stand-alone firm if it does not belong to a group and does not rely on significant shareholder debt financing. We assume that every firm which has a German natural person as global ultimate owner is a stand-alone firm. This consideration is based on the fact that given a natural person is the ultimate owner, the firm may actually stand-alone or, if this is not the case, the firm is part of a group which can be tax consolidated. In both cases, the thin capitalization rule does not apply. In sample 1 where only firms with ownership information have been included, 66 firms with net interest expenses above 1 million euro are entitled to the stand-alone-escape clause. 11

13 Tax Group Stand-Alone-Clause: In case the firm had a profit and loss agreement in place and the global ultimate owner is a German company, we considered the firm as being a part of a tax consolidated group and thus assumed that the thin capitalization rule does not apply. 36 firms of the 168 treated firms in sample 1 are based on our modeling entitled to tax group stand-alone-escape clause. EBITDA-Clause: In case a firm has a ratio of interest expenses to for tax purposes adjusted EBITDA below 30 % the thin capitalization rule does not apply. We construct the for tax purposes adjusted EBITDA by adding back the depreciation allowance, the net interest result and the provisions which are not allowed for tax purposes to the before tax profit. For this escape clause 56 firms of the treated 168 firms in sample 1 are entitled. Figure 1: Debt Ratio for Treatment and Control Group: Common Trend Assumption Debt Ratio 67,0% 66,0% 65,0% 64,0% 63,0% 62,0% 61,0% 60,0% 59,0% 58,0% 57,0% 56,0% Control Group Treatment Group Notes: Sample 1, for further descriptive statistics see table A1 in the appendix. Source: DAFNE firm database, own calculations. Before we turn to the description of the variables for our second and third research question, we plot the development of the mean debt ratio for treatment and control group in sample 1 to check whether treatment and control group exhibit a common trend before the reform (Figure 1). Two things are obvious in the figure. First, between 2004 and 2006 both groups show a similar evolution of the mean debt ratio, thus we conclude that the 12

14 common trend assumption for treatment and control group is satisfied. 19 Secondly, already the graphic analysis shows a strong decline in the debt ratio for the treated firms indicating that the thin capitalization rule affected firm s financial structure. For our second research question, the outcome variable is the change of the capital stock between 2008 and 2006, scaled by the capital stock in We refer to this variable as the investment quota. As control variables we include into the estimation equation two important factors of the investment quota, the change in the tax rate on business income and the change of the log. firm age. Additionally, due to the observed difference between treatment and control group with respect to these variables, we further control for firm size and firms share of tangible assets with their values in To analyze whether the interest barrier broadens the tax base, we use the change in the profit before taxes scaled by the book value of total assets between 2008 and As for investment, we include the change in the business tax rate, the change in the log firm age as well as the 2006 levels of firm size and firms tangibility. 5. Results We start the presentation of our results with the analysis of the impact of the thin capitalization rule in Germany on the debt ratio of our treatment and control group as defined in the last section. Then we will present our findings on the impact of this thin capitalization rule on investment. In the last subsection, we report the results on the question whether the interest barrier has broadened the German tax base. Debt Ratio In the first specification we regress the Treatment * After variable as well as the control variables on the change of the debt ratio between 2008 and The result for this specification is reported in specification (1) in Table 2. The estimated coefficient for the impact of the thin capitalization rule (line Treatment * After in Table 2) states that firms, which are affected by the thin capitalization rule have, on average, reduced their debt ratio 19 The figure for the development of the debt ratio in sample 2 and 3 are very similar to the one shown above. Also for investment and probability the common trend is clearly observed in all samples. Figures are not shown but are available upon request. 13

15 by about 2.5 percentage points. Given this adjustment, at least part of these firms are able to fully deduct their interest expenses because a lower debt ratio implies lower interest expenses. In all cases where the interest expenses after adjustment fall below the exemption limit, these expenses are fully deductible. Table 2: Regression Analysis: Change in Debt Ratio for Treatment and Control Group (1) (2) (3) (4) (Treatment * After) ** -.028** *** *** (0.010) (0.013) (0.011) (0.018) (Treatment * After * EBITDA ESCAPE) (0.021) (After * EBITDA ESCAPE) (0.009) (Treatment * After * Stand Alone 0.046** (0.020) ( After * Stand Alone) (0.009) (log Firm size) 0.075*** (0.027) (0.279) (log Firm age) * * * (0.007) (0.009) (0.007) (0.007) (Tangibility) (0.071) (0.482) (Tax rate business income) (0.091) (0.143) (0.099) (0.104) (EBITDA/Total assets) *** ** *** *** (0.081) (0.110) (0.083) (0.082) L2. log Firm size (0.006) (0.007) L2. Tangibility ** (0.013) (0.014) (After) (0.007) (0.026) (0.059) (0.063) Observations R Notes: Dependent variable is the two year change of the debt ratio, defined as liabilities to total assets. indicates the change between 2008 and L2 indicates twice lagged variables. The treatment group consists of firms with net interest expenses between 1,2 and 1,5 million EUR in The control group includes firms with net interest expenses between 0,5 and 0,8 million euro. In the specification (2) we use an instrumental variable approach for (log Firm size) and (Tangibility). The F-test (Shea s Partial R 2) amounts to (0.008) for (log. Firm size) and to (0.015) for (Tangibility). As excluded instrument we use the twice lagged level of the variables. Stars (***/**/*) indicate significance at the 1\%/5\%/10\% levels. Source: Dafne firm database, years 2006 and 2008, own calculations 14

16 Since it is very likely that firm size and firm s tangibility are endogenous in the equation for the debt ratio because they are affected by changes in the firms finance structure, the estimated coefficients are probably biased. In specification (2), we thus used an instrumental variable approach to deal with the endogeneity of the two variables. As excluded instruments we used the twice lagged level of the variables. In specification (2), the coefficient of interest is and thus slightly higher than in specification (1). However, the test statistics, shown in the notes of table 2, indicate that these instruments are weak. In specification (3) we have used the twice lagged levels of firm size and firm s tangibility as explanatory variables. In this case, the coefficient for the Treatment * After variable increases further to Since also the efficiency of the estimates rises we stick to this specification of the control variables for all further estimations. Although the control variables are no longer subject to endogeneity issues, it is still very likely that the estimated coefficient of interest is downward biased as our treatment group only accounts for the exemption limit but not for other escape clauses. In order to consider these escape clauses in the estimation, we add suitable interaction terms to the estimation equation. The first term interacts the Treatment * After variable with a dummy for the stand-alone-escape clause (including the tax group stand-alone escape clause). The second term interacts the Treatment * After variable with a dummy for the EBITDA escape clause. Two further interaction term are formed in the same way using the dummy After instead of the dummy Treatment * After. The result for specification (4) Is in line with our expectations: the coefficient of the variable Treatment * After increases considerably (to 0.054). However, given that the escape clauses are used by the firms which are entitled to them, we would also expect that the two respective interaction terms with Treatment * After were significant. However, this is the case only for the stand-alone-escape clause. The coefficient for this interaction term amounts to 0.046, which indicates that firms which had interest expenses above 1 million euro in 2006 and which are entitled to the stand escape clause do not reduce their debt ratio. This is in line with our expectations. Contrary to our expectation, the point estimate for the interaction between Treatment * After and the dummy indicating whether a firm is not affected by the interest barrier because the firms interest expenses amount to less than 30 % of EBITDA 30 % is negative (although insignificant). We see two possible explanations for this result. First, our measure of EBITDA, adjusted for tax purposes, is not very accurate. We might therefore actually not a capture 15

17 the firms we intend to capture. Second, the EBITDA escape clause could be a very risky as interest rate shocks or demand shocks affect this ratio strongly. It seems that this option is rarely used. Since we cannot assess the relevance of these causes, we leave the question to future research. Further, since the estimated coefficients for the EBITDA escape clause variables are insignificant, we leave them out in the following specifications. Further, as the interaction term for the stand-alone escape clause with the After variable is also not significant and the reaction for firms which are entitled to the standalone-escape clause is not statistically different from zero ( ), we redefine our treatment group. In the following, our treatment group only consists of firms which have net interest expenses between 1,200 and 1,500 thousand euro and which are not entitled to the stand-alone-escape clause. In the control group, we include firms which are entitled to the stand-alone-escape clause or which show net interest expenses between 500 and 800 thousand euros. Specification (1) in table 3 shows the results for the baseline specification with the redefined treatment and control groups. The coefficient for the Treatment * After variable amounts to and indicates that treated firms decreased their debt ratio by about 6.1 percentage points. In the following specification, we analyze whether firms exploit the exemption limit of 1 million euro by splitting up their assets. Since the most preferable way to divide the firm s assets is to incorporate a subsidiary to which a part of the assets is carried over we used for this analyses the change of the number of the firms subsidiaries between 2008 and We construct a dummy variable which is one in case the number of subsidiaries increased and zero else. This dummy variable is interacted with the Treatment * After as well as the After variable. In case, firms used the exemption limit to avoid the application of the thin capitalization rule, the interaction term with Treatment * After should be positive, since these firms do not have to reduce their debt ratio if their net interest expenses are below the 1 million euro threshold. Further, the effect of the treated firms should increase as without controlling for firms who split up the estimated coefficient was downward biased. The results of specification (2) in table 3 support the view that firms used the exemption threshold. The coefficient for the Treatment * After variable increases to and the one for the interaction term between the dummy indicating whether the number of subsidiaries 16

18 has increased and Treatment * After amounts to Thus, firms which split up their assets did not reduce their debt ratio between 2008 and Table 3: Regression Analysis: Change in debt ratio (1) (2) (Treatment * After) *** *** (0.016) (0.018) (Treatment * After) * Dummy( Subsidiaries) 0.075*** (0.029) ( After) * Dummy( Subsidiaries) * (0.010) (log Firm age) * ** (0.007) (0.007) (Tax rate business income) (0.103) (0.102) L2. log Firm size * (0.005) (0.005) L2. Tangibility 0.025* 0.025** (0.013) (0.013) (After) * * (0.054) (0.053) Observation R Notes: Depend variable is the two year change of the debt ratio, defined as liabilities to total assets. indicates the change between 2008 and L2 indicates twice lagged variables. The treatment group consists of firms with net interest expenses between 1,2 and 1,5 million EUR in 2006 and which are not entitled to the stand-alone-escape clause. The control group includes firms with net interest expenses between 0,5 and 0,8 million euro and these firms which are entitled to the stand-alone-escape clause. Stars (***/**/*) indicate significance at the 1\%/5\%/10\% levels. Source: DAFNE firm database, years 2006 and 2008, own calculations. Before we analyze the impact of the interest barrier on firms investment decisions, we come back to our two other samples. For both samples, we cannot model the escape clauses besides for the firms already included in sample 1. We expect thus the coefficient of interest is more in line with the specification (3) in table 1 where the escape clauses have not been considered. For both samples, we estimate the baseline specification and the specification for the splitting up hypothesis. The results are shown in table A-2 in the appendix. For the second sample, where all firms with an income statement have been included the coefficient for the Treatment * After variable amounts to and is thus slightly higher than the estimates in specification (3) for sample 1, which is not surprising given that we account for the a part of the escape clause. In the second specification for the split up behavior the 17

19 coefficient for Treatment * After increase to Also for the second sample, the interaction term for the treated firms which increased their number of subsidiaries is positive. This confirms our analysis for sample 1. For the third sample, the results are weaker given that only for less than half of the firms the escape clauses can be modeled. However, the results for the baseline specification are only slightly smaller than those from specification (3) with sample 1. Our results so far confirm the studies by Overesch and Wamser (2011) and Buettner et al. (2008) who find that firms react to thin capitalization rules. The results are however not directly comparable since the thin capitalization rules they analyzed are based on equity debt ratios whereas the one analyzed in this paper is based on the exemption threshold. Further, our analysis has shown that firms use different strategies to avoid the interest barrier. Firms which have the possibility to split up do so with the result that the thin capitalization rule does not affect their finance behavior compared to firms who cannot split and have to reduce their debt ratio by several percentage points. Investment quota In this section we report our results for the effect of the introduction of the interest barrier on firms investment behavior. The investment behavior could be affected due to two different channels. The first channel depends on the substitutability of debt and equity. In case equity and debt are no perfect substitutes we expect that the reduction of the debt ratio affects the investment behavior negatively due to presence of finance frictions. The second channel why investment could be affected due to the interest barrier is related to international profit shifting. In case firms shift profits, the tax burden on investment is lower compared to firms which do not shift profits. If now a thin capitalization rule is imposed, which prevents firms to shift profits the tax burden on investment increases. Thus investment should be adversely affected. The negative effect should be higher the higher the gain of profit shifting before the introduction of the interest barrier was. Noteworthy, the effect on investment is not negative per se but depends whether the thin capitalization rule is introduced in a high or in low tax country. 18

20 Table 4: Change Investment Quota for Treatment and Control Group (1) (2) (3) (4) (5) (Treatment * After) ** ** (0.047) (0.050) (0.055) (0.015) (0.047) (Treatment * After) * Dummy( Subsidiaries) * (0.081) ( After) * Dummy( Subsidiaries) (0.040) (TRD) 0.415* (0.238) (Treatment * TRD) 1.245*** (0.307) (After * TRD) *** (0.181) (Treatment * After * TRD) 1.721*** (0.497) Cash flow to total assets (0.121) (Treatment * Cash flow to total assets) 0.854*** (0.279) (Tax rate business income) (0.481) (0.525) (0.480) (0.476) d2_firmage * (0.023) (0.025) (0.023) (0.010) (0.000) L2. log Firm size (0.021) (0.023) (0.021) (0.011) (0.020) L2. Tangibility *** *** *** *** *** (0.060) (0.060) (0.060) (0.014) (0.060) (log sales) 0.120*** (0.043) (After) 0.402* ** 0.386*** 0.425** (0.221) (0.241) (0.222) (0.114) (0.213) Observation R Notes: Depend variable is the change of the capital stock between 2008 and 2006 scaled by the capital stock in indicates the change between 2008 and L2 indicates twice lagged variables. The treatment group consists of firms with net interest expenses between 1,2 and 1,5 million EUR in 2006 and which are not entitled to the stand-alone-escape clause. The control group includes firms with net interest expenses between 0,5 and 0,8 million euro and these firms which are entitled to the stand-alone-escape clause. Stars (***/**/*) indicate significance at the 1\%/5\%/10\% levels. Source: DAFNE firm database, years 2006 and 2008, own calculations. Given these considerations, we start our analysis with the redefined treatment and control. The dependent variable is the change in the capital stock between 2008 and 2006 scaled by the capital stock in We refer to this as the investment rate. As control variables we include the change of the tax rate on business income, the change of the firm age as well as the level of firm size and firms tangibility in 2006 to control for differences 19

21 between treatment and control group. The results for the baseline specification are reported in table (4), specification (1); the coefficient of the Treatment * After variable is negative but insignificant. This indicates that treated firms had on average sufficient equity to finance its investment project. In the second specification, we also included the growth rate of sales between 2008 and 2006 (measured as difference of the log of total sales in thousand euros). The coefficient for the Treatment * After variable remains insignificant. Before, we come to the specification where we control for internal cash and the tax rate differential between host and foreign country to analyze the hypothesis outlined above, we analyze whether firms for which we think they split up their assets indeed do this. To analyze this question, we interact the dummy indicating whether the number of subsidiaries has increased with the Treatment * After variable. Specification (3) in table 5 reports the results. The interaction term is indeed negative and amounts to This indicates that firms split up to use the exemption threshold in order to avoid the interest barrier. In specification (4) and (5) we test whether there is effect heterogeneity in the investment behavior of firms. In specification (4) we analyze whether the internal finance decision affects the investment behavior. We do so by interacting firms cash flow of 2006 and 2008 scaled by the book value of total assets with the Treatment * After and the After variable. In case investment depends on internal cash flow, we expect a positive sign for the first interaction term. The findings show (specification (4), table 5) that indeed the investment effect depends on the internally available cash flow for the treated firms. The coefficient for the interaction term amounts to and thus indicates that almost every free available euro is used for investment by the treated firms. In the last specification, we checked whether the investment behavior of the treated firms depends on tax rate differential (TRD), which is defined as the difference between the tax rate in the host and the tax rate in the foreign country. We analyze this question by including the respective interactions between the tax rate differential and difference-indifference dummies. The equation we estimate has the form of equation (3): The dependent variable is the change in investment. The Treatment * After variable captures the treatment effect not associated with the tax rate differential. The variable of interest is given by 20